TechPrufer
Get started
DeepMind
DeepMind logoDeepMind · Feature trail

AI-powered vulnerability detection and remediation

1 post · Jul 2026 – Sep 2026

What this trail covers

Gemini models are enhanced for live voice interactions, including improved speech-to-text and speech-to-speech translation. Gemini 3.8 Live and 3.8 Live Extended Thinking offer advancements in near real-time reasoning for voice agents, with improved conversational intelligence, fluid dialogue, visual grounding, and multi-step reasoning. These models support over 97 languages mid-conversation, execute tools and API calls in the background, and provide live progress narration for complex tasks. The Gemini 3.5 Flash Cyber model is introduced, a lightweight cybersecurity model built on top of 3.5 Flash and fine-tuned to find, validate, and patch vulnerabilities quickly and efficiently. It offers a cost-efficient and highly capable alternative to large, costly cybersecurity models and is integrated into CodeMender for automated vulnerability discovery and remediation. It has demonstrated competitive performance against larger models on benchmarks like CyberGym and Big Sleep, and is being used internally at Google for production commit scanning and vulnerability research.

All DeepMind trails
  1. Proactive cyber defense for governments and enterprisesThis post introduces the Fairwind Program, which integrates Gemini 3.8 Flash Cyber with the CodeMender harness to enable autonomous vulnerability finding and fixing. It highlights the ability to generate verified, deployment-ready patches in minutes within a secure cloud environment, contrasting this with traditional manual fixes that take weeks. The program targets governments, critical infrastructure operators, and core technology platforms, emphasizing responsible use through strict operational standards.
  2. Introducing Gemini 3.8 Flash and 3.8 Flash CyberIntroduces Gemini 3.8 Flash, a new model with improved reasoning and coding capabilities, outperforming larger frontier models on long-horizon software engineering tasks and specialized domains like finance and legal analysis. It achieves this through more diligent execution, including extra reasoning steps and iterative tool calls. Gemini 3.8 Flash Cyber is a new cybersecurity-focused model that demonstrates frontier-level performance in autonomous vulnerability discovery on CyberGym and a comprehensive internal benchmark across 20 languages. It also shows strong automated patching capabilities on the CWE-Bench benchmark, achieving a pass@1 of 47.2% at a significantly lower cost than leading frontier models. The post highlights real-world impact, with Gemini 3.8 Flash Cyber producing 2.6x more correct patches for Chrome vulnerabilities than comparable commercial models and finding a critical Google Cloud vulnerability in under 2 hours. The model also shows improved prompt injection robustness.
  3. Introducing Gemini 3.5 Flash CyberIntroduces Gemini 3.5 Flash Cyber, a lightweight cybersecurity model fine-tuned for vulnerability detection, validation, and patching. It leverages the performance and efficiency of Gemini 3.5 Flash to provide a cost-effective alternative to larger cybersecurity models. The post details its integration into CodeMender, enabling multiple invocations for deeper code analysis and improved vulnerability discovery. Benchmark results on CyberGym, Big Sleep, and Chrome's production commit scanning pipeline show competitive and often superior performance compared to mainline Flash models and other large language models. It highlights real-world application within Google, including finding remote code execution vulnerabilities and memory corruption issues, and discusses its scalability for securing large codebases.